The Software Layer the Grid Has Been Missing

Distributed energy resources need real-time coordination. AI-based grid balancing platforms are finally mature enough to do that job.

The power grid was designed around a simple premise: centralized generation, one-way power flow, predictable demand. Large coal and gas plants produced electricity. Transmission lines moved it to substations. Distribution networks carried it to homes and businesses. Utilities managed supply to match demand. The physics were well understood. The operational models were mature.

That model started breaking down somewhere around 2018, when distributed solar penetration in markets like California and Texas crossed the threshold where utility-scale forecasting tools started failing with uncomfortable regularity. The problem is not that solar is unreliable — it is that the grid's control architecture was never designed for resources scattered across millions of rooftops and utility-scale farms, each responding independently to local conditions.

Why Traditional SCADA Systems Cannot Keep Up

SCADA — the supervisory control and data acquisition systems that grid operators have used for decades — works well for a grid with a few hundred controllable generation assets. It was not designed for a grid with tens of thousands of distributed resources that each have their own local controllers and optimization logic. The communication protocols are too slow. The update intervals are too long. And the models underlying the dispatch decisions were built in an era when demand was the variable and supply was the constant. Now both sides of the equation are variable simultaneously.

The consequence is that grid operators face increasing frequency of what they call "non-consequential" balancing events — periods where supply and demand are technically in balance, but only because operators made expensive real-time adjustments using gas peakers or curtailed renewable generation. Both responses cost money. Curtailment wastes clean electrons. Peaker dispatch emits carbon. These are the symptoms of a coordination problem, not a supply problem.

We have seen data from grid operators showing that 15 to 20 percent of renewable curtailment in high-penetration markets is not caused by transmission constraints or lack of storage — it is caused by insufficient real-time forecasting and dispatch coordination. That is a software problem with a software solution.

What AI-Based Dispatch Actually Does

Prism Grid — our Series B portfolio company in this space — builds what they describe as a "distributed energy resource management system" with machine learning models running continuously across their client networks. The core technical differentiation is in the forecasting layer: rather than relying on regional aggregate demand and generation forecasts, their system builds fine-grained probabilistic forecasts at the feeder level, updated on five-minute intervals.

That granularity matters because grid balancing events are often localized. A thunderstorm that reduces solar output across one distribution feeder simultaneously increases cooling demand in the same area — and the two effects compound in a way that aggregate forecasting models smooth over. Feeder-level models catch it. They dispatch storage assets on that specific feeder rather than signaling curtailment at the regional level.

The results in deployed systems are measurable. In a 2025 study covering three utility deployments, Prism Grid's platform reduced curtailment events by 31 percent and peaker dispatch frequency by 22 percent compared to the prior 12-month baseline. Those are not small numbers for a utility that runs on thin margins and answers to a state commission about renewable integration costs.

The Regulatory and Integration Challenge

The technical side of this problem is largely solved. The harder challenge is integration with utility IT systems and regulatory structures that were built for a different world. Most utilities run their operational technology on infrastructure that is 10 to 20 years old. Connecting a modern distributed resource management platform to that infrastructure requires significant systems integration work — and utilities are, reasonably, risk-averse about making changes to operational systems that affect real-time grid reliability.

The regulatory side is even more complicated. Distributed energy resource dispatch crosses the jurisdictional boundary between distribution (state PUC territory) and bulk power (FERC territory) in ways that existing interconnection agreements do not cleanly address. FERC Order 2222, finalized in 2020, took a step toward enabling distributed resources to participate in wholesale markets — but implementation has been slow and inconsistent across regional transmission organizations.

We expect this to resolve over the next three to five years. The economics are compelling enough that utilities and regulators are motivated to work through the integration complexity. The renewable penetration levels that make grid coordination software necessary are not optional — they are coming regardless of the software readiness, because the generation projects are already permitted and under construction.

Why Software Scales Differently Than Hardware

One of the reasons we like this space as an investment category is the margin structure. Grid hardware — transformers, storage systems, transmission equipment — has capital-intensive economics with high variable costs per unit of deployment. Software platforms, once the core product is built and validated, scale with relatively modest incremental cost per additional megawatt under management.

The competitive dynamics favor early movers who can demonstrate reliability at scale. Grid operators will not switch platforms based on a 10 percent cost savings if the incumbent has two years of flawless operational history. Getting the first large utility deployment right is worth more than any feature list. That is where we focus our evaluation: not just the technology, but the team's ability to operate reliably in a zero-tolerance environment.

The grid has needed this software layer for a decade. It is finally here.